Professional Documents
Culture Documents
L θ
α V 1
replacements
γ 0
L (N)
M
−1
−2
δ
−3
−40 −20 0 20 40 60
α (deg)
Figure 1: Illustration of longitudinal aircraft entities.
Figure 2: Typical lift force vs angle of attack relationship.
x2 would be successfully brought to the origin. This
recursive procedure is repeated until the actual control pitching moment. This will affect the angle of attack
variable u is reached after n steps, whereby a stabilizing to which the lift force is strongly related. Note that
control law, u = φn−1 (x), has been constructed. this is exactly the same assumption which is needed
for dynamic inversion to be applied to aircraft control
Along with the control law, a Lyapunov function is con-
[6]. Expressing L and M in terms of their aerodynamic
structed, proving the stability of the closed loop system.
coefficients, we have that
L = q̄SCL (α)
(6)
3 Aircraft model M = q̄Sc̄Cm (α, q, δ)
where q̄ = 12 ρV 2 is the dynamic pressure, ρ is the air
In this paper only control about the longitudinal axis density, S is the wing platform area, and c̄ is the mean
is considered. With this restriction, the equations of aerodynamic chord.
motion describing the aircraft take the form [10]
1
V̇ = (−D + FT cos α − mg sin γ) (2) 4 Design for flight path angle control
m
1
γ̇ = (L + FT sin α − mg cos γ) (3) 4.1 Deriving the control law
mV
θ̇ = q (4) The system description consists of Equations (3)-(5),
where the flight path angle γ is to be controlled. Given
1
q̇ = (M + FT ZT P ) (5) the reference value, γref , the angle of attack at steady
Iy state, α0 , is defined through γ̇ = 0, see Equation (3).
In the actual implementation, α0 is computed on-line
The state variables are V = airspeed, γ = flight path
and updated at each time step, using current values of
angle, θ = pitch angle, and q = pitch rate. α = θ − γ is
q̄, δ 1 , and FT . Although this pragmatic approach to de-
the angle of attack, FT is the engine thrust force, which
termining α0 is not guaranteed to converge, successful
contributes to the pitching moment due to a thrust
simulations provide an alibi.
point offset, ZT P . The aerodynamical effects on the
aircraft are captured by the lift and drag forces, L and In Equation (3), the γ dependence of the gravitational
D, and the pitching moment, M . See Figure 1, where term plays an insignificant role. Therefore we will only
δ represents the deflections of the control surfaces that consider its contribution at the equilibrium, and assume
are at our disposal. the dynamics to be
The controlled variables are V and γ. In the imple- 1
γ̇ = (L(α) + FT sin α − mg cos γref )
mented controller, evaluated in Section 5, airspeed con- mV (7)
trol is handled separately using results from [7]. Our , ϕ(α − α0 )
control design will therefore focus solely on Equations
(3)-(5). where ϕ(0) = 0. Using the sign properties of its com-
ponents, L(α), see Figure 2 for a typical example, and
To meet the structural demand in Section 2, we will FT sin α, we conclude that
neglect the dependence of the lift force on the pitch
αϕ(α) > 0, α 6= 0 (8)
rate and the control surface deflections. The principal
function of the control surfaces to produce an angular 1 In computing α0 , the lift force dependence on δ is included.
for all α of practical interest2 . This is a vital property an extra degree of freedom, where F is required to be
for the backstepping control design to follow which will positive definite and radially growing. This extension
be done in three steps. of the backstepping procedure is due to [5]. Hence,
c2 2 1 2
Step 1: First introduce the control error V2 = z + z + F (ξ), c2 > 0
2 1 2 2
z1 = γ − γref We compute its time derivative to find a new stabilizing
function, qdes .
whose dynamics are given by
V̇2 = c2 z1 ϕ(ξ) + z2 (q + c1 ϕ(ξ)) + F 0 (ξ)(−ϕ(ξ) + q)
ż1 = ϕ(α − α0 ) (9)
= (c2 z1 + c1 z2 − F 0 (ξ))ϕ(ξ) + (z2 + F 0 (ξ))q
considering a constant reference value. Now use the
control Lyapunov function (clf)
Although it may not be transparent, we can again find
1 a stabilizing function independent of ϕ. Choosing
V1 = z12
2
qdes = −c3 z2 , c3 > 0 (11)
to determine a stabilizing function, θdes , considering θ 0
F (ξ) = c4 ϕ(ξ), F (0) = 0, c4 > 0 (12)
as the control input of Equation (9).
where (12) is an implicit but perfectly valid choice of
V̇1 = z1 ϕ(α − α0 ) F , yields
= z1 ϕ(θ − z1 − γref − α0 )
= z1 ϕ(−(1 + c1 )z1 + θ + c1 z1 − γref − α0 ) V̇2 = (c2 z1 + (c1 − c3 c4 )z2 )ϕ(ξ) − c4 ϕ(ξ)2 − c3 z22
= z1 ϕ(−(1 + c1 )z1 ) < 0, z1 6= 0
Selecting
is achieved by selecting
c2 = −(1 + c1 )(c1 − c3 c4 ), c 3 c4 > c 1 (13)
θdes = −c1 z1 + γref + α0 , c1 > −1 (10)
gives the negative definiteness of V̇2 that we want since
The fact that c1 = 0 is a valid choice means that γ
feedback is not necessary for the sake of stabilization. V̇2 = (c1 − c3 c4 )ξϕ(ξ) − c4 ϕ(ξ)2 − c3 z22 < 0
However, it provides an extra degree of freedom for tun-
ing the closed loop performance. for all z1 , z2 6= 0, again using (8). The benefit of using
the extra term F (ξ) shows up in Equation (13). F (ξ) ≡
Step 2: Continue by introducing the deviation from 0 leads to c2 = −(1+c1 )c1 > 0 and the severe restriction
the virtual control law (10). c1 < 0 implying positive feedback from z1 .
5 Application −30
craft. The model was originally released for the GAR- True dq/dt
0.6 Model excl. E
TEUR robust flight control challenge [8]. Model incl. E
0.4
For the simulations, a flight case where the aircraft is 0.2
in level flight at Mach 0.3 and at a height of 5000 ft was
0
chosen. Actuator and sensor dynamics, which were not
considered in the preceding design, were included in the −0.2
simulations. Also, a constant pitch coefficient error of
−0.4
−0.030 was added to Cm . The simulation results are
reproduced in Figure 3. −0.6
−0.8
The top picture displays the flight path angle time re- 0 5 10 15 20
time (s)
sponse to a staircase reference signal. Also plotted is
the time response of the desired linear closed loop sys-
tem, i.e., the control law applied to the linearization of Figure 3: From top to bottom: γ time response compared
(3)-(5). We see that for small changes in the reference, with the designed linear response, α time re-
the simulated response stays very close to the linear sponse, taileron deflection, and finally the true
one. For the large step in the reference, occurring af- q̇ compared with Equation (21), with and with-
ter 5 seconds, the simulated response has a longer rise out the model error estimate E.
time. This has two main causes. First, we see that the the control variable, we have shown a linear control law
control surfaces saturate at the beginning of the step, to be globally stabilizing despite the nonlinear nature
thus not producing the desired pitching moment. This of the lift force with respect to the angle of attack. The
consequently slows down the step response. Second, fact that the control law is linear makes it easy to tune
the angle of attack becomes very large during the ma- given a linearization of the open loop system. Using this
neuver. When α reaches its peak after 7 seconds, the the locally linear closed loop behavior can be selected.
true lift force value is significantly smaller than what
the linear model predicts, see Figure 2. In agreement Simulations using a realistic aircraft model including
with Equation (3), this also leads to slower γ dynamics. actuator and sensor dynamics, and a constant pitch
coefficient error proved the control law to be robust
In the bottom picture, the benefit of estimating the against the approximations made during the design.
model error E can be seen. The dotted curve, rep-
resenting the complete model of the angular pitch ac- A comparison with feedback linearization showed the
celeration introduced in Equation (21), initially has a potential benefits of using backstepping. The backstep-
bias error compared to the true q̇. This error is quickly ping control law is computationally much simpler, and
estimated whereafter the curve closely follows the solid is globally stabilizing while feedback linearization yields
curve, representing the true q̇. The latter was computed a singularity in the control law around the stall angle.
by differentiating q sensor data.
8 Acknowledgment
6 Backstepping vs feedback linearization
The authors would like to acknowledge the Defence
Having derived the globally stabilizing control law (18) Evaluation and Research Agency (DERA) of the United
for (3)-(5) using backstepping, it is rewarding to make Kingdom for supplying the HIRM model.
a comparison with a feedback linearizing control de-
sign. Again using z1 = γ − γref , feedback linearization
proceeds by defining References
[1] R.P. Brent. Algorithms for Minimization Without
ż1 = ϕ(α − α0 ) = z2 Derivatives. Prentice-Hall, 1973.
ż2 = ϕ0 (α − α0 )(q − z2 ) = z3 [2] G.E. Forsythe, M.A. Malcolm, and C.B. Moler.
ż3 = ϕ00 (α − α0 )(q − z2 )2 + ϕ0 (α − α0 )(u − z3 ) = ũ Computer Methods for Mathematical Computations.
Prentice-Hall, 1976.
T
Selecting ũ = − k1 k2 k3 z1 z2 z3 clearly [3] A.J. Krener and A. Isidori. Linearization by out-
renders the closed loop system linear in the z coor- put injection and nonlinear observers. Systems & Con-
dinates. Solving for the angular acceleration u to be trol Letters, 3:47–52, June 1983.
produced, defined as in (15), we get
[4] M. Krstić, I. Kanellakopoulos, and P. Kokotović.
ũ − ϕ00 (α − α0 )(q − z2 )2 Nonlinear and Adaptive Control Design. John Wiley &
u = z3 + (24) Sons, 1995.
ϕ0 (α − α0 )
[5] M. Krstić and P.V. Kokotović. Lean backstepping
Two things are worth noting about this expression. design for a jet engine compressor model. In Proceedings
First, it depends not only ϕ, but also on its first and of the 4th IEEE Conference on Control Applications,
second derivatives. Recalling its definition from Equa- pages 1047–1052, 1995.
tion (7), this means that the lift force L must be well [6] S.H. Lane and R.F. Stengel. Flight control de-
known in order to accurately compute u. In the back- sign using non-linear inverse dynamics. Automatica,
stepping design, we relied on L only through its zero 24(4):471–483, 1988.
crossing and its sign switching property (8). Second, ϕ0
is in the denominator of the expression above, implying [7] T. Larsson. Linear quadratic design and µ-
that the control law has a singularity where ϕ0 = 0. analysis of a fighter stability augmentation system.
This occurs around the stall angle, where the lift force Master’s thesis, Linköpings universitet, 1997.
no longer increases with α, see Figure 2. [8] J-F. Magni, S. Bennani, and J. Terlouw, editors.
Robust Flight Control - A Design Challenge. Springer,
1997.
7 Conclusions [9] Jean-Jaques E. Slotine and Weiping Li. Applied
Nonlinear Contol. Prentice Hall, 1991.
In this paper we have applied backstepping to flight [10] B.L. Stevens and F.L. Lewis. Aircraft Control and
path angle control. Regarding the pitching moment as Simulation. John Wiley & Sons, 1992.